From: AAAI Technical Report SS-03-01. Compilation copyright © 2003, AAAI (www.aaai.org). All rights reserved.
Working with a Knowledge Management Tool in a Domain Context
Tang-Ho Lê1 and Luc Lamontagne2
1
Computer Science Department
Université de Moncton, 165 Massey Ave, Moncton, (N.B.)
CANADA, E1A 3E9
2
Départment d’informatique et de recherche opérationelle
Université de Montréal, C.P. 6128, Succ. Centre-Ville, Montréal (Québec)
CANADA, H3C 3J7
letangho@umoncton.ca
lamontal@iro.umontreal.ca
Abstract
In this paper, we provide an overview of our software tool
to exploit and interchange procedural knowledge
represented as networks of semi-structured units. First, we
introduce the notion of Procedural Knowledge Hierarchy;
then we present the modeling of Procedural Knowledge by
our software. We claim that the “bottom-up” approach,
that can be carried out by this tool, is appropriate to gather
new candidate terms for the construction of a new domain
ontology. An educational project, conducted with our tool,
is briefly reported.
and specify knowledge being procedural in nature, and to
combine this with declarative knowledge available on the
web. In this paper, we report on our work to address some
of these issues. First, we emphasize the essential role of
procedural knowledge in the knowledge interchange, then
we discuss our software tool used to describe procedural
knowledge in a semi-structured form. Afterwards, we
report on some experiences using this tool in the practical
context of an e-learning project.
2 Procedural Knowledge
1 Introduction
Current research on the semantic web has emerged from
the most pressing problems of information overload on the
Web. This phenomenon is natural as the Web represents a
virtual medium in which people, with different cultural
and knowledge background, can disseminate documents
and interchange ideas, products, etc.
However
exploitation of this information is complicated by its
abundance. For instance, the evolution of terminologies
and their semantic is a source of ambiguity. Several new
words occur daily and some terms even change their
meaning when used in different domains. For example,
the term “home page” did not exist before and the term
“navigating” now has a new meaning when used in the
context of the Web which differs from its nautical origins.
The research community has been conducting for
several years intensive research to devise formal methods
to understand and to determine declarative knowledge
based on ontologies. Many research groups are formed
and contribute results to this issue; for example, several
proposals were made for ontology languages and ontology
builder such as DAML, Protege 2000, Ontolingua, etc.
But there is also a need to better represent procedural
knowledge which is still poorly understood. To our
knowledge, fewer research efforts were devoted to model
Copyright © 2002, American Association for Artificial Intelligence
(www.aaai.org). All rights reserved.
To progress toward a semantic web in a large sense, the
classification of concepts into a giant hierarchy, with the
terms being assigned according to the community’s
consensus, is just one half of the solution. This half is
based on the knowledge about knowledge, while the other
half is, after having access to the right knowledge in this
giant hierarchy, to know how to perform some tasks to
produce the thing abstracted by this knowledge. Put in
other words, we want to know how to create the product P
after having the concept of P. This refers to various
knowledge to be tackled. The first (categorize and
classify) suggest “the ontology” that belongs to the
declarative knowledge while the second (how to perform
some tasks) is the procedural knowledge.
The procedural knowledge is at the core of the
knowledge that we interchange daily. To interchange
efficiently, first, a consensus about the elements of
discourse is needed, and then, some knowledge about how
to achieve or better perform our tasks is required. The
world (or an application domain) can not be solely
conceived as static, its evolution being partially made
possible through human actions.
Our knowledge
encompasses the knowledge about these actions. This
kind of knowledge is essentially what we want to know
(know how) in addition to the “know what” in our daily
learning. For instance, in the web page domain, we firstly
need the definition of terms such as hyperlink, site
structure, etc. but we also need to know “how to make
links” and “how to navigate in the web site”.
3 Procedural Knowledge Hierarchy
In the design of a Knowledge Management (KM) system,
one has to cope with the granularity problem. If we want
to efficiently transfer the desired knowledge to knowledge
workers, what will be its suitable volume? To provide
answers to this question, we believe that one must take
into account the user’s cognitive capability.
This
cognitive limitation, which can hinder the transfer of
knowledge, is difficult to determine for two reasons :
§ Because the knowledge is not well defined yet.
Fundamentally, is the knowledge to be transfered
declarative or procedural in nature? The former
concerns concept definitions or factual information
while the latter is essentially actions to perform some
tasks (as an algorithm). In the both cases, calling them
“knowledge” might confuse the receiver.
§ Because of the lack of a clear separation between
information and knowledge. Consequently, instead of
transfering the right level of knowledge, we may
collect some information of all types. In the worse
case, the receiver of this “knowledge”, cannot consume
it, has the impression that he/she always lacks the
desired knowledge and wants to know more. In other
cases, one put them in the “Knowledge Pool” to deal
with later.
For the transfer of knowledge, some research results
already available from Intelligent Tutoring Systems (ITS)
are applicable. In such systems, a knowledge object (or
knowledge unit) corresponds roughly to a teaching subject.
Several related subjects form a lesson (what we call a
Knowledge Network - KN) that one can teach in a lecture
(during about one hour).
Moreover, it has been
determined that the teaching of procedural knowledge will
be more efficient if accompanied by demonstrations (Lê,
Gauthier, and Frasson 1998) which can be realized by
animation files (e.g. by Macromedia Flash) or video films.
Even in the context of ontological negotiation as
proposed in (van Elst and Abecker, 2002), one must first
frame out the underlying knowledge. To do this, a
cognitive analysis is necessary. From the light of ITS, we
can design a KN as a task hierarchy consisting of just a
small number of levels (from one to three levels is ideal).
That is, the KN is named as a global task, and its content
can be detailed with some Knowledge Units as primitive
tasks which are arranged as a sequence or as a hierarchy.
While the ontology construction is based on the entities of
the external word, the procedural knowledge relies on our
epistemological states of mind. This distinction leads to
these two important consequences. First, the ontology
construction can apply the principles of object oriented
design with class definition (a template to classify
instances) in which the emphasis is put on the view of
entities (expressed by nouns) and the relation between
super and sub-classes as generalization and specialization.
However, the actions are designed as relations between
two different class entities; for example, a “Car” being
driven by a “Person” (Noy and McGuinness 2001).
Second, for procedure knowledge, the importance relies on
the description of actions or the achievement of tasks. In
order to make understandable this knowledge,
pedagogical principles can be applied, e.g. one can
specify the prerequisite (condition) for each knowledge
progression step. Thus, as a result, one produces also a
hierarchy of domain knowledge concepts which are a mix
of entity concepts and procedural concepts. The latter can
be expressed by verbs and by nouns derived from verbs
(not like the entity nouns); for example, “navigation in the
web site”.
4 Procedural Knowledge Management Tool
The design of a KM system refers essentially to the
modeling of knowledge so that it is understandable to the
users, or at least, to provide them with the conditions for
an adequate awareness. As for many KM systems, we
make use of frames to model knowledge. To clarify the
terminology, we use the term “Knowledge Unit” (KU) to
designate some knowledge about an entity or about a
process. Thus, we distinguish two kinds of units: "Static"
KU and "How-to" KU. A static KU contains concept
definitions, labels, facts and information related to the
underlying domain. The names of these KU form (or are
derived from) the domain ontology. How-to units are
task-oriented. They contain the procedures to follow
through (actions or tasks) and refers to static KU when
necessary. The description of a How-to KU, in natural
language, focuses on a limited scope, and its attributes
specify the underlying context, a situation or some
conditions. This context allows users to become aware of
the conditions to understand it. A How-to KU can also
give some references (links) to other available documents.
Knowledge units always relate to other KUs, linked
together as Knowledge Networks (KN). Such networks
provide a complete view of the knowledge related to a
specific topic.
For additionnal information on the
structure of the knowledge units and networks, we refer
the reader to [Le and Nguyen 2001].
In the current version, in addition to the subtask links,
the KN contains three other kinds of links: The workflow
links reflect the order between KUs in an activity, a project
or an organization. The pedagogical links connect
prerequisite KU for understanding the actual KU. The
required knowledge is called prerequisite knowledge
which is usually some Static KUs that provide users with
explanations on the domain ontology (e.g. concept
definitions). The logical links provide information on the
logical relationships between KUs (e.g. a KU being
deduced from another one). Logical links are provided to
support the description of reasoning schemes. A global
KU index is maintained by the system. Existing KUs can
be reused by integrating them into newly created KNs.
Our tool (KMT) can be used as an environment in
which we are mostly interested on how new terms,
emerging from the worker’s creative knowledge, are
integrated within a knowledge hierarchy. This is of
interest because an ontology reflects this hierarchy of
knowledge [Chandrasekaran, Josephson, and Benjamins
1998]. In other words, we would like to intervene upon
the emergence of a new term. This task can be seen as the
“bottom-up” approach to construct an ontology. This
approach involves the capturing of the knowledge in
evolution to keep trace of the domain progresses.
In our KMT environment, the content of the How-to
units is considered as a corpus and each new significant
term with its explicit meaning is recorded in a Static KU.
When the number of KN reach a manageable size, a
grouping task is necessary. We call it the sub-domain
promotion (it seems like the promotion of an index level
in the B-tree indexing). The knowledge engineer should
name a sub-domain by a generic category term which
conceptually covers all the underneath KNs. These
category terms progressively form the domain knowledge
hierarchy. The method proposed in (Biebow and Szulman
2000) can help in this task. The knowledge hierarchy is
then updated in the domain attribute for these KNs.
the knowledge creation among them. In addition, the
leader could provide them with some internet resource
links to redirect them about some relevant public
documents. At this point, we feel that the result of the
project has been sustained by the use of this tool.
5 Preliminary Experiments
References
8 Conclusion
Languages, tools and methods for building domain
ontologies are increasingly perceived as means to achieve
the semantic web. However, research efforts are still
required to devise adequate formalisms for the
management of procedural knowledge assets.
In this
paper, we provided an overview of our software tool to
exploit and interchange procedural knowledge represented
as networks of semi-structured units. We claim that the
“bottom-up” approach, that can be carried out with this
tool, is appropriate to gather new terms for the
construction of a new domain ontology which will be
candidates for the discussion/selection by the underlying
community. An educational project, conducted with our
tool, is briefly reported. For future work, we intend to
include some high-level categories which would remain
stable for most domains and upon which new declarative
knowledge could be integrated to the knowledge networks.
A concept hierarchy inspired from the university structure
(with faculties, schools, departments) is being considered
as a candidate model.
1
During the summer of 2001, we conducted a project to
improve the self-learning abilities of forty workers. One
of the problem we had to face was to coordinate the
authoring of knowledge resources by contributors residing
in three different cities. The KMT was used to set up the
content of a distance learning course that teaches Web
page creation using the FrontPage software (17 lessons, 65
KUs and 35 Flash files for demonstration).
The
distributed version of the KMT was also used for
knowledge interchange between the participants in our
collaborative work. At the beginning, the team leader had
created a template KN including some units representing
the tasks to be accomplished by each member. During
their work, the members gradually refined their own KU
with some subtask units and new findings. The updated
knowledge networks were made available to the project
leader and all members. The KN continually grew up until
a final state was reached. So the sequence of changes
brought to the KN was informative for following the
evolution of the project’s knowledge. Mutual review and
comparison of the member’s experience also stimulated
1
This project was financially supported by the ministry of
Human Resources Development Canada.
Biebow, B. and Szulman, S. 2000 Une approche terminologique
pour categoriser les concepts d'une ontologie, in Ingénierie des
connaissances : Évolutions récentes et nouveaux défis, Eyrolles,
Paris, France.
Chandrasekaran, B., Josephson, J. R. and Benjamins V. R. 1998
The Ontology of Tasks and Methods, in Proceedings of
KAW’98, Gaines, B. and Musen, M. (eds.), Banff, Canada.
Noy, N. and McGuinness, D. L. 2001 Ontology Development
101: A Guide to Creating Your First Ontology. 2001, Technical
Report: SMI-2001-0880, Stanford Medical Informatics.
Lê, T.H. and Nguyen, T.H. 2001 Toward an Universal
Knowledge Management: Using Ontology Approach with
methods and tools in Proceedings of the 5th World
Multiconference on Systemics, Cybernetics and Informatics
(SCI-2001), 79-83, Orlando, USA.
Lê, T.H., Gauthier, G. and Frasson, C. 1998 The Process of
Planning for an Intelligent Tutoring System, in Proceeding of
The Fourth World Congress On Expert Systems, 709–714,
Mexico City, Mexico.
van Elst, L. and Abecker, A. 2002. Domain Ontology Agents in
Distributed
Organizational
Memories,
in
Knowledge
Management and Organizational Memories, Kluwer Academic
Publishers, USA.